Welcome to The Manifest: Why Declarative Agents Matter

After a decade of building on Microsoft 365, shipping Graph APIs, and helping customers integrate with Teams, SharePoint, and Outlook, I’ve seen waves come and go: webhooks, bots, adaptive cards, message extensions. Some stuck. Some didn’t. But nothing has excited me like what’s happening right now with declarative agents.

So I built this blog. Welcome to The Manifest.

Why This Blog Exists

Every week, I talk to developers and IT pros who are trying to figure out the Microsoft 365 Copilot extensibility story. And every week, I hear the same thing: “I know Copilot is powerful, but I don’t know where to start extending it.” Or worse: “I tried, got lost in the docs, and gave up.”

That’s a problem. Not because the documentation isn’t good (it is), but because the landscape is moving fast, and what developers really need is someone who’s been in the trenches, connecting the dots, showing the patterns, and being honest about what works and what doesn’t.

That’s what The Manifest is for. Practical, opinionated, code-heavy content about building on Microsoft 365 Copilot. No fluff. No corporate-speak. Just a developer talking to developers.

The Hidden Gem: Declarative Agents

Here’s what I believe most people are sleeping on: declarative agents are the most underrated feature in the entire Microsoft 365 extensibility stack.

Think about it. You can build an AI-powered agent that:

Has a custom persona with specific instructions and tone
Is grounded in your organization’s SharePoint documents, emails, and Teams messages
Can call your APIs through API plugins
Can execute code through MCP servers
Can generate images, interpret data, and search the web
Runs inside Microsoft 365 Copilot with enterprise security and compliance built in

And you do all of this with zero infrastructure. No servers to manage. No databases to provision. No auth flows to implement from scratch. You write a JSON manifest, some instructions in plain English, and connect your data sources. That’s it.

{
  "$schema": "https://developer.microsoft.com/json-schemas/copilot/declarative-agent/v1.6/schema.json",
  "version": "v1.6",
  "name": "The Manifest Agent",
  "description": "Your guide to M365 Copilot extensibility",
  "instructions": "You are a helpful assistant that answers questions about building declarative agents for Microsoft 365 Copilot. Be practical, include code examples, and always reference official documentation."
}

That’s a working agent. I’m not oversimplifying; that’s genuinely how you start. The Agents Toolkit scaffolds it, you can test it immediately in M365 Copilot, and one click deploys it to your entire organization.

Want to ground it in your company’s SharePoint documents and restrict web search to your own domain? Add a capabilities array:

{
  "capabilities": [
    {
      "name": "OneDriveAndSharePoint",
      "items_by_url": [
        { "url": "https://contoso.sharepoint.com/sites/policies" }
      ]
    },
    {
      "name": "WebSearch",
      "sites": [
        { "url": "https://contoso.com" }
      ]
    }
  ]
}

Two JSON blocks, zero infrastructure. Your agent now pulls answers from internal documents and your public website, all running inside Microsoft 365 Copilot with enterprise security and compliance built in.

What You’ll Find Here

This blog covers everything you need to build on Microsoft 365 Copilot extensibility: from your first declarative agent to production-ready deployments. Expect practical, opinionated content with real scenarios, honest takes, and code you can actually use.

Every post follows the same philosophy: start with a real scenario, explain why it matters, show you how to do it, and give you code you can copy.

Who Am I?

I’m Sébastien Levert, a Principal Product Manager at Microsoft working on the Microsoft 365 Copilot ecosystem. I’ve spent years building developer tools and platforms, from the Microsoft Graph Toolkit to the Teams Toolkit to what’s now the Agents Toolkit. I’ve seen what works at scale, and I’ve seen what breaks in production.

This blog is my way of sharing everything I’ve learned: the wins, the gotchas, and the patterns that actually hold up when real users start asking real questions.

Bookmark this blog, try building an agent, and when you break something, come back and tell me about it.

Why This Blog Exists

The Hidden Gem: Declarative Agents

What You’ll Find Here

Who Am I?

Resources